On Wed, 2025-02-26 at 14:08 +0100, Ingo Molnar wrote:
>
> * Benjamin Berg <benjamin@xxxxxxxxxxxxxxxx> wrote:
>
> > From: Benjamin Berg <benjamin.berg@xxxxxxxxx>
> >
> > The init_task instance of struct task_struct is statically allocated and
> > may not contain the full FP state for userspace. As such, limit the copy
> > to the valid area of init_task and fill the rest with zero.
> >
> > Note that the FP state is only needed for userspace, and as such it is
> > entirely reasonable for init_task to not contain parts of it.
> >
> > Signed-off-by: Benjamin Berg <benjamin.berg@xxxxxxxxx>
> > Fixes: 5aaeb5c01c5b ("x86/fpu, sched: Introduce CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT and use it on x86")
> > ---
> > arch/x86/kernel/process.c | 10 +++++++++-
> > 1 file changed, 9 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
> > index f63f8fd00a91..1be45fe70cad 100644
> > --- a/arch/x86/kernel/process.c
> > +++ b/arch/x86/kernel/process.c
> > @@ -92,7 +92,15 @@ EXPORT_PER_CPU_SYMBOL_GPL(__tss_limit_invalid);
> > */
> > int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
> > {
> > - memcpy(dst, src, arch_task_struct_size);
> > + /* init_task is not dynamically sized (incomplete FPU state) */
> > + if (unlikely(src == &init_task)) {
> > + memcpy(dst, src, sizeof(init_task));
> > + memset((void *)dst + sizeof(init_task), 0,
> > + arch_task_struct_size - sizeof(init_task));
> > + } else {
> > + memcpy(dst, src, arch_task_struct_size);
>
> Note that this patch, while it still applies cleanly, crashes/hangs the
> x86-64 defconfig kernel bootup in the early boot phase in a KVM guest
> bootup.
Oh, outch. It seems that arch_task_struct_size can actually become
smaller than sizeof(init_task) if the CPU does not have certain
features.
See fpu__init_task_struct_size, which does:
int task_size = sizeof(struct task_struct);
task_size -= sizeof(current->thread.fpu.__fpstate.regs);
task_size += fpu_kernel_cfg.default_size;
I'll submit a new version of the patch and then also switch to use
memcpy_and_pad.
Benjamin
